summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht/src/dht-helper.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht/src/dht-helper.c')
-rw-r--r--xlators/cluster/dht/src/dht-helper.c326
1 files changed, 243 insertions, 83 deletions
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 055ea7cb0..311a48112 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -18,6 +18,28 @@
#include "xlator.h"
#include "dht-common.h"
+static inline int
+dht_inode_ctx_set1 (xlator_t *this, inode_t *inode, xlator_t *subvol)
+{
+ uint64_t tmp_subvol = 0;
+
+ tmp_subvol = (long)subvol;
+ return inode_ctx_set1 (inode, this, &tmp_subvol);
+}
+
+int
+dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol)
+{
+ int ret = -1;
+ uint64_t tmp_subvol = 0;
+
+ ret = inode_ctx_get1 (inode, this, &tmp_subvol);
+ if (tmp_subvol && subvol)
+ *subvol = (xlator_t *)tmp_subvol;
+
+ return ret;
+}
+
int
dht_frame_return (call_frame_t *frame)
@@ -40,6 +62,43 @@ dht_frame_return (call_frame_t *frame)
}
+static uint64_t
+dht_bits_for (uint64_t num)
+{
+ uint64_t bits = 0, ctrl = 1;
+
+ while (ctrl < num) {
+ ctrl *= 2;
+ bits ++;
+ }
+
+ return bits;
+}
+
+/*
+ * A slightly "updated" version of the algorithm described in the commit log
+ * is used here.
+ *
+ * The only enhancement is that:
+ *
+ * - The number of bits used by the backend filesystem for HUGE d_off which
+ * is described as 63, and
+ * - The number of bits used by the d_off presented by the transformation
+ * upwards which is described as 64, are both made "configurable."
+ */
+
+
+#define BACKEND_D_OFF_BITS 63
+#define PRESENT_D_OFF_BITS 63
+
+#define ONE 1ULL
+#define MASK (~0ULL)
+#define PRESENT_MASK (MASK >> (64 - PRESENT_D_OFF_BITS))
+#define BACKEND_MASK (MASK >> (64 - BACKEND_D_OFF_BITS))
+
+#define TOP_BIT (ONE << (PRESENT_D_OFF_BITS - 1))
+#define SHIFT_BITS (max (0, (BACKEND_D_OFF_BITS - PRESENT_D_OFF_BITS + 1)))
+
int
dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
{
@@ -47,6 +106,9 @@ dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
int cnt = 0;
int max = 0;
uint64_t y = 0;
+ uint64_t hi_mask = 0;
+ uint64_t off_mask = 0;
+ int max_bits = 0;
if (x == ((uint64_t) -1)) {
y = (uint64_t) -1;
@@ -60,7 +122,23 @@ dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
max = conf->subvolume_cnt;
cnt = dht_subvol_cnt (this, subvol);
- y = ((x * max) + cnt);
+ if (max == 1) {
+ y = x;
+ goto out;
+ }
+
+ max_bits = dht_bits_for (max);
+
+ hi_mask = ~(PRESENT_MASK >> (max_bits + 1));
+
+ if (x & hi_mask) {
+ /* HUGE d_off */
+ off_mask = MASK << max_bits;
+ y = TOP_BIT | ((x >> SHIFT_BITS) & off_mask) | cnt;
+ } else {
+ /* small d_off */
+ y = ((x * max) + cnt);
+ }
out:
if (y_p)
@@ -135,16 +213,38 @@ dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
int max = 0;
uint64_t x = 0;
xlator_t *subvol = 0;
+ int max_bits = 0;
+ uint64_t off_mask = 0;
+ uint64_t host_mask = 0;
if (!this->private)
- goto out;
+ return -1;
conf = this->private;
max = conf->subvolume_cnt;
- cnt = y % max;
- x = y / max;
+ if (max == 1) {
+ x = y;
+ cnt = 0;
+ goto out;
+ }
+
+ if (y & TOP_BIT) {
+ /* HUGE d_off */
+ max_bits = dht_bits_for (max);
+ off_mask = (MASK << max_bits);
+ host_mask = ~(off_mask);
+
+ x = ((y & ~TOP_BIT) & off_mask) << SHIFT_BITS;
+
+ cnt = y & host_mask;
+ } else {
+ /* small d_off */
+ cnt = y % max;
+ x = y / max;
+ }
+out:
subvol = conf->subvolumes[cnt];
if (subvol_p)
@@ -153,7 +253,6 @@ dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
if (x_p)
*x_p = x;
-out:
return 0;
}
@@ -263,20 +362,6 @@ out:
return local;
}
-
-char *
-basestr (const char *str)
-{
- char *basestr = NULL;
-
- basestr = strrchr (str, '/');
- if (basestr)
- basestr ++;
-
- return basestr;
-}
-
-
xlator_t *
dht_first_up_subvol (xlator_t *this)
{
@@ -428,7 +513,36 @@ out:
return next;
}
+/* This func wraps around, if prev is actually the last subvol.
+ */
+xlator_t *
+dht_subvol_next_available (xlator_t *this, xlator_t *prev)
+{
+ dht_conf_t *conf = NULL;
+ int i = 0;
+ xlator_t *next = NULL;
+ conf = this->private;
+ if (!conf)
+ goto out;
+
+ for (i = 0; i < conf->subvolume_cnt; i++) {
+ if (conf->subvolumes[i] == prev) {
+ /* if prev is last in conf->subvolumes, then wrap
+ * around.
+ */
+ if ((i + 1) < conf->subvolume_cnt) {
+ next = conf->subvolumes[i + 1];
+ } else {
+ next = conf->subvolumes[0];
+ }
+ break;
+ }
+ }
+
+out:
+ return next;
+}
int
dht_subvol_cnt (xlator_t *this, xlator_t *subvol)
{
@@ -620,20 +734,36 @@ dht_migration_complete_check_task (void *data)
call_frame_t *frame = NULL;
loc_t tmp_loc = {0,};
char *path = NULL;
+ dht_conf_t *conf = NULL;
+ inode_t *inode = NULL;
+ fd_t *iter_fd = NULL;
+ uint64_t tmp_subvol = 0;
+ int open_failed = 0;
this = THIS;
frame = data;
local = frame->local;
+ conf = this->private;
src_node = local->cached_subvol;
- /* getxattr on cached_subvol for 'linkto' value */
- if (!local->loc.inode)
+ if (!local->loc.inode && !local->fd)
+ goto out;
+
+ inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+ /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+ * as root:root. If a fd is already open, access check wont be done*/
+
+ if (!local->loc.inode) {
ret = syncop_fgetxattr (src_node, local->fd, &dict,
- DHT_LINKFILE_KEY);
- else
+ conf->link_xattr_name);
+ } else {
+ SYNCTASK_SETID (0, 0);
ret = syncop_getxattr (src_node, &local->loc, &dict,
- DHT_LINKFILE_KEY);
+ conf->link_xattr_name);
+ SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+ }
if (!ret)
dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
@@ -684,10 +814,7 @@ dht_migration_complete_check_task (void *data)
/* update inode ctx (the layout) */
dht_layout_unref (this, local->layout);
- if (!local->loc.inode)
- ret = dht_layout_preset (this, dst_node, local->fd->inode);
- else
- ret = dht_layout_preset (this, dst_node, local->loc.inode);
+ ret = dht_layout_preset (this, dst_node, inode);
if (ret != 0) {
gf_log (this->name, GF_LOG_DEBUG,
"%s: could not set preset layout for subvol %s",
@@ -705,10 +832,7 @@ dht_migration_complete_check_task (void *data)
goto out;
}
- if (!local->loc.inode)
- ret = dht_layout_set (this, local->fd->inode, layout);
- else
- ret = dht_layout_set (this, local->loc.inode, layout);
+ ret = dht_layout_set (this, inode, layout);
if (ret) {
gf_log (this->name, GF_LOG_ERROR,
"%s: failed to set the new layout",
@@ -719,42 +843,46 @@ dht_migration_complete_check_task (void *data)
local->cached_subvol = dst_node;
ret = 0;
- if (!local->fd)
+ /* once we detect the migration complete, the inode-ctx2 is no more
+ required.. delete the ctx and also, it means, open() already
+ done on all the fd of inode */
+ ret = inode_ctx_reset1 (inode, this, &tmp_subvol);
+ if (tmp_subvol)
goto out;
- /* once we detect the migration complete, the fd-ctx is no more
- required.. delete the ctx, and do one extra 'fd_unref' for open fd */
- ret = fd_ctx_del (local->fd, this, NULL);
- if (!ret) {
- fd_unref (local->fd);
- ret = 0;
+ if (list_empty (&inode->fd_list))
goto out;
- }
- /* if 'local->fd' (ie, fd based operation), send a 'open()' on
- destination if not already done */
- if (local->loc.inode) {
- ret = syncop_open (dst_node, &local->loc,
- local->fd->flags, local->fd);
- } else {
- tmp_loc.inode = local->fd->inode;
- inode_path (local->fd->inode, NULL, &path);
- if (path)
- tmp_loc.path = path;
- ret = syncop_open (dst_node, &tmp_loc,
- local->fd->flags, local->fd);
- GF_FREE (path);
+ /* perform open as root:root. There is window between linkfile
+ * creation(root:root) and setattr with the correct uid/gid
+ */
+ SYNCTASK_SETID(0, 0);
+
+ /* perform 'open()' on all the fd's present on the inode */
+ tmp_loc.inode = inode;
+ inode_path (inode, NULL, &path);
+ if (path)
+ tmp_loc.path = path;
+ list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+ if (fd_is_anonymous (iter_fd))
+ continue;
+ ret = syncop_open (dst_node, &tmp_loc,
+ iter_fd->flags, iter_fd);
+ if (ret == -1) {
+ gf_log (this->name, GF_LOG_ERROR, "failed to open "
+ "the fd (%p, flags=0%o) on file %s @ %s",
+ iter_fd, iter_fd->flags, path, dst_node->name);
+ open_failed = 1;
+ }
}
- if (ret == -1) {
- gf_log (this->name, GF_LOG_ERROR,
- "%s: failed to send open() on target file at %s",
- local->loc.path, dst_node->name);
+ GF_FREE (path);
+
+ SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+ if (open_failed) {
+ ret = -1;
goto out;
}
-
- /* need this unref for the fd on src_node */
- fd_unref (local->fd);
ret = 0;
out:
@@ -798,20 +926,34 @@ dht_rebalance_inprogress_task (void *data)
char *path = NULL;
struct iatt stbuf = {0,};
loc_t tmp_loc = {0,};
+ dht_conf_t *conf = NULL;
+ inode_t *inode = NULL;
+ fd_t *iter_fd = NULL;
+ int open_failed = 0;
this = THIS;
frame = data;
local = frame->local;
+ conf = this->private;
src_node = local->cached_subvol;
- /* getxattr on cached_subvol for 'linkto' value */
- if (local->loc.inode)
+ if (!local->loc.inode && !local->fd)
+ goto out;
+
+ inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+ /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+ * as root:root. If a fd is already open, access check wont be done*/
+ if (local->loc.inode) {
+ SYNCTASK_SETID (0, 0);
ret = syncop_getxattr (src_node, &local->loc, &dict,
- DHT_LINKFILE_KEY);
- else
+ conf->link_xattr_name);
+ SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+ } else {
ret = syncop_fgetxattr (src_node, local->fd, &dict,
- DHT_LINKFILE_KEY);
+ conf->link_xattr_name);
+ }
if (ret) {
gf_log (this->name, GF_LOG_ERROR,
@@ -853,33 +995,46 @@ dht_rebalance_inprogress_task (void *data)
ret = 0;
- if (!local->fd)
- goto out;
+ if (list_empty (&inode->fd_list))
+ goto done;
+
+ /* perform open as root:root. There is window between linkfile
+ * creation(root:root) and setattr with the correct uid/gid
+ */
+ SYNCTASK_SETID (0, 0);
+
+ tmp_loc.inode = inode;
+ inode_path (inode, NULL, &path);
+ if (path)
+ tmp_loc.path = path;
+
+ list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+ if (fd_is_anonymous (iter_fd))
+ continue;
- if (local->loc.inode) {
- ret = syncop_open (dst_node, &local->loc,
- local->fd->flags, local->fd);
- } else {
- tmp_loc.inode = local->fd->inode;
- inode_path (local->fd->inode, NULL, &path);
- if (path)
- tmp_loc.path = path;
ret = syncop_open (dst_node, &tmp_loc,
- local->fd->flags, local->fd);
- GF_FREE (path);
+ iter_fd->flags, iter_fd);
+ if (ret == -1) {
+ gf_log (this->name, GF_LOG_ERROR, "failed to send open "
+ "the fd (%p, flags=0%o) on file %s @ %s",
+ iter_fd, iter_fd->flags, path, dst_node->name);
+ open_failed = 1;
+ }
}
+ GF_FREE (path);
- if (ret == -1) {
- gf_log (this->name, GF_LOG_ERROR,
- "%s: failed to send open() on target file at %s",
- local->loc.path, dst_node->name);
+ SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+
+ if (open_failed) {
+ ret = -1;
goto out;
}
- ret = fd_ctx_set (local->fd, this, (uint64_t)(long)dst_node);
+done:
+ ret = dht_inode_ctx_set1 (this, inode, dst_node);
if (ret) {
gf_log (this->name, GF_LOG_ERROR,
- "%s: failed to set fd-ctx target file at %s",
+ "%s: failed to set inode-ctx target file at %s",
local->loc.path, dst_node->name);
goto out;
}
@@ -931,6 +1086,9 @@ dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
dht_stat_time_t *time = 0;
int ret = -1;
+ GF_VALIDATE_OR_GOTO (this->name, stat, out);
+ GF_VALIDATE_OR_GOTO (this->name, inode, out);
+
ret = dht_inode_ctx_get (inode, this, &ctx);
if (ret) {
@@ -948,6 +1106,8 @@ dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
DHT_UPDATE_TIME(time->atime, time->atime_nsec,
stat->ia_atime, stat->ia_atime_nsec, inode, post);
+ ret = dht_inode_ctx_set (inode, this, ctx);
+out:
return 0;
}
generated by cgit (git 2.34.1) at 2026-01-05 19:50:54 +0000